The Calculus of Computation

Transcription

1 The Calculus of Computation

2 Aaron R. Bradley Zohar Manna The Calculus of Computation Decision Procedures with Applications to Verification With 60 Figures 123

3 Authors Aaron R. Bradley Zohar Manna Gates Building, Room 481 Stanford University Stanford, CA USA Library of Congress Control Number: ACM Computing Classification (1998): B.8, D.1, D.2, E.1, F.1, F.3, F.4, G.2, I.1, I.2 ISBN Springer Berlin Heidelberg New York This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilm or in any other way, and storage in data banks. Duplication of this publication or parts thereof is permitted only under the provisions of the German Copyright Law of September 9, 1965, in its current version, and permission for use must always be obtained from Springer. Violations are liable for prosecution under the German Copyright Law. Springer is a part of Springer Science+Business Media springer.com Springer-Verlag Berlin Heidelberg 2007 The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. Typesetting by the authors Production: LE-TEX Jelonek, Schmidt & Vöckler GbR, Leipzig Cover design: KünkelLopka Werbeagentur, Heidelberg Printed on acid-free paper 45/3180/YL

4 To my wife, Sarah A.R.B. To my grandchildren, Itai Maya Ori Z.M.

5 Preface Logic is the calculus of computation. Forty-five years ago, John McCarthy predicted in A Basis for a Mathematical Theory of Computation that the relationship between computation and mathematical logic will be as fruitful in the next century as that between analysis and physics in the last. The field of computational logic emerged over the past few decades in partial fulfillment of that vision. Focusing on producing efficient and powerful algorithms for deciding the satisfiability of formulae in logical theories and fragments, it continues to push the frontiers of general computer science. This book is about computational logic and its applications to program verification. Program verification is the task of analyzing the correctness of a program. It encompasses the formal specification of what a program should do and the formal proof that the program meets this specification. The reasoning power that computational logic offers revolutionized the field of verification. Ongoing research will make verification standard practice in software and hardware engineering in the next few decades. This acceptance into everyday engineering cannot come too soon: software and hardware are becoming ever more ubiquitous and thus ever more the source of failure. We wrote this book with an undergraduate and beginning graduate audience in mind. However, any computer scientist or engineer who would like to enter the field of computational logic or apply its products should find this book useful. Content The book has two parts. Part I, Foundations, presents first-order logic, induction, and program verification. The methods are general. For example, Chapter 2 presents a complete proof system for first-order logic, while Chapter 5 describes a relatively complete verification methodology. Part II, Algorithmic Reasoning, focuses on specialized algorithms for reasoning about fragments of first-order logic and for deducing facts about programs. Part II trades generality for decidability and efficiency.

6 VIII Preface The first three chapters of Part I introduce first-order logic. Chapters 1 and 2 begin our presentation with a review of propositional and predicate logic. Much of the material will be familiar to the reader who previously studied logic. However, Chapter 3 on first-order theories will be new to many readers. It axiomatically defines the various first-order theories and fragments that we study and apply throughout the rest of the book. Chapter 4 reviews induction, introducing some forms of induction that may be new to the reader. Induction provides the mathematical basis for analyzing program correctness. Chapter 5 turns to the primary motivating application of computational logic in this book, the task of verifying programs. It discusses specification, in which the programmer formalizes in logic the (sometimes surprisingly vague) understanding that he has about what functions should do; partial correctness, which requires proving that a program or function meets a given specification if it halts; and total correctness, which requires proving additionally that a program or function always halts. The presentation uses the simple programming language pi and is supported by the verifying compiler πvc (see The πvc System, below, for more information on πvc). Chapter 6 suggests strategies for applying the verification methodology. Part II on Algorithmic Reasoning begins in Chapter 7 with quantifierelimination methods for limited integer and rational arithmetic. It describes an algorithm for reducing a quantified formula in integer or rational arithmetic to an equivalent formula without quantifiers. Chapter 8 begins a sequence of chapters on decision procedures for quantifier-free and other fragments of theories. These fragments of first-order theories are interesting for three reasons. First, they are sometimes decidable when the full theory is not (see Chapters 9, 10, and 11). Second, they are sometimes efficiently decidable when the full theory is not (compare Chapters 7 and 8). Finally, they are often useful; for example, proving the verification conditions that arise in the examples of Chapters 5 and 6 requires just the fragments of theories studied in Chapters The simplex method for linear programming is presented in Chapter 8 as a decision procedure for deciding satisfiability in rational and real arithmetic without multiplication. Chapters 9 and 11 turn to decision procedures for non-arithmetical theories. Chapter 9 discusses the classic congruence closure algorithm for equality with uninterpreted functions and extends it to reason about data structures like lists, trees, and arrays. These decision procedures are for quantifier-free fragments only. Chapter 11 presents decision procedures for larger fragments of theories that formalize array-like data structures. Decision procedures are most useful when they are combined. For example, in program verification one must reason about arithmetic and data structures simultaneously. Chapter 10 presents the Nelson-Oppen method for combining decision procedures for quantifier-free fragments. The decision procedures of Chapters 8, 9, and 11 are all combinable using the Nelson-Oppen method. Chapter 12 presents a methodology for constructing invariant generation procedures. These procedures reason inductively about programs to aid in

7 Preface IX 1 4 5, Fig The chapter dependency graph 11 Verification Decision procedures verification. They relieve some of the burden on the programmer to provide program annotations for verification purposes. For now, developing a static analysis is one of the easiest ways of bringing formal methods into general usage, as a typical static analysis requires little or no input from the programmer. The chapter presents a general methodology and two instances of the method for deducing arithmetical properties of programs. Finally, Chapter 13 suggests directions for further reading and research. Teaching This book can be used in various ways and taught at multiple levels. Figure 0.1 presents a dependency graph for the chapters. There are two main tracks: the verification track, which focuses on Chapters 1 4, 5, 6, and 12; and the decision procedures track, which focuses on Chapters 1 4 and Within the decision procedures track, the reader can focus on the quantifier-free decision procedures track, which skips Chapters 7 and 11. The reader interested in quickly obtaining an understanding of modern combination decision procedures would prefer this final track. We have annotated several sections with a to indicate that they provide additional depth that is unnecessary for understanding subsequent material. Additionally, all proofs may be skipped without preventing a general understanding of the material. Each chapter ends with a set of exercises. Some require just a mechanical understanding of the material, while others require a conceptual understanding or ask the reader to think beyond what is presented in the book. These latter exercises are annotated with a. For certain audiences, additional exercises might include implementing decision procedures or invariant generation procedures and exploring certain topics in greater depth (see Chapter 13). In our courses, we assign program verification exercises from Chapters 5 and 6 throughout the term to give students time to develop this important skill. Learning to verify programs is about as difficult for students as learning

8 X Preface to program in the first place. Specifying and verifying programs also strengthens the students facility with logic. Bibliographic Remarks Each chapter ends with a section entitled Bibliographic Remarks in which we attempt to provide a brief account of the historical context and development of the chapter s material. We have undoubtedly missed some important contributions, for which we apologize. We welcome corrections, comments, and historical anecdotes. The πvc System We implemented a verifying compiler called πvc to accompany this text. It allows users to write and verify annotated programs in the pi programming language. The system and a set of examples, including the programs listed in this book, are available for download from arbrad/pivc. We plan to update this website regularly and welcome readers comments, questions, and suggestions about πvc and the text. Acknowledgments This material is based upon work supported by the National Science Foundation under Grant Nos. CSR and CNS and by Navy/ONR contract N Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation or the Navy/ONR. The first author received additional support from a Sang Samuel Wang Stanford Graduate Fellowship. We thank the following people for their comments throughout the writing of this book: Miquel Bertran, Andrew Bradley, Susan Bradley, Chang- Seo Park, Caryn Sedloff, Henny Sipma, Matteo Slanina, Sarah Solter, Fabio Somenzi, Tomás Uribe, the students of CS156, and Alfred Hofmann and the reviewers and editors at Springer. Their suggestions helped us to improve the presentation substantially. Remaining errors and shortcomings are our responsibility. Stanford University, June 2007 Aaron R. Bradley Zohar Manna

9 Contents Part I Foundations 1 Propositional Logic Syntax Semantics Satisfiability and Validity Truth Tables Semantic Arguments Equivalence and Implication Substitution Normal Forms Decision Procedures for Satisfiability Simple Decision Procedures Reconsidering the Truth-Table Method Conversion to an Equisatisfiable Formula in CNF The Resolution Procedure DPLL Summary Bibliographic Remarks Exercises First-Order Logic Syntax Semantics Satisfiability and Validity Substitution Safe Substitution Schema Substitution Normal Forms Decidability and Complexity Satisfiability as a Formal Language

10 XII Contents Decidability Complexity Meta-Theorems of First-Order Logic Simplifying the Language of FOL Semantic Argument Proof Rules Soundness and Completeness Additional Theorems Summary Bibliographic Remarks Exercises First-Order Theories First-Order Theories Equality Natural Numbers and Integers Peano Arithmetic Presburger Arithmetic Theory of Integers Rationals and Reals Theory of Reals Theory of Rationals Recursive Data Structures Arrays Survey of Decidability and Complexity Combination Theories Summary Bibliographic Remarks Exercises Induction Stepwise Induction Complete Induction Well-Founded Induction Structural Induction Summary Bibliographic Remarks Exercises Program Correctness: Mechanics pi: A Simple Imperative Language The Language Program Annotations Partial Correctness Basic Paths: Loops Basic Paths: Function Calls

11 Contents XIII Program States Verification Conditions P-Invariant and P-Inductive Total Correctness Summary Bibliographic Remarks Exercises Program Correctness: Strategies Developing Inductive Annotations Basic Facts The Precondition Method A Strategy Extended Example: QuickSort Partial Correctness Total Correctness Summary Bibliographic Remarks Exercises Part II Algorithmic Reasoning 7 Quantified Linear Arithmetic Quantifier Elimination Quantifier Elimination A Simplification Quantifier Elimination over Integers Augmented Theory of Integers Cooper s Method A Symmetric Elimination Eliminating Blocks of Quantifiers Solving Divides Constraints Quantifier Elimination over Rationals Ferrante and Rackoff s Method Complexity Summary Bibliographic Remarks Exercises Quantifier-Free Linear Arithmetic Decision Procedures for Quantifier-Free Fragments Preliminary Concepts and Notation Linear Programs The Simplex Method

12 XIV Contents From M to M Vertex Traversal Complexity Summary Bibliographic Remarks Exercises Quantifier-Free Equality and Data Structures Theory of Equality Congruence Closure Algorithm Relations Congruence Closure Algorithm Congruence Closure with DAGs Directed Acyclic Graphs Basic Operations Congruence Closure Algorithm Decision Procedure for T E -Satisfiability Complexity Recursive Data Structures Arrays Summary Bibliographic Remarks Exercises Combining Decision Procedures Combining Decision Procedures Nelson-Oppen Method: Nondeterministic Version Phase 1: Variable Abstraction Phase 2: Guess and Check Practical Efficiency Nelson-Oppen Method: Deterministic Version Convex Theories Phase 2: Equality Propagation Equality Propagation: Implementation Correctness of the Nelson-Oppen Method Complexity Summary Bibliographic Remarks Exercises Arrays Arrays with Uninterpreted Indices Array Property Fragment Decision Procedure Integer-Indexed Arrays

13 Contents XV Array Property Fragment Decision Procedure Hashtables Hashtable Property Fragment Decision Procedure Larger Fragments Summary Bibliographic Remarks Exercises Invariant Generation Invariant Generation Weakest Precondition and Strongest Postcondition General Definitions of wp and sp Static Analysis Abstraction Interval Analysis Karr s Analysis Standard Notation and Concepts Summary Bibliographic Remarks Exercises Further Reading References Index